Re: [squid-users] Tweaking Squid for speed, not max requests

On Fri, Aug 31, 2001, Brett Lymn wrote:
> According to Brett Lymn:
> >
> >According to Adrian Chadd:
> >>
> >>Yes, but you can only complete one file operation at a time.
> >>
> >
> >Ah - yes, that is true.
> >
>
> Dammit following up to myself...
>
> But the original post was implying (unless I read it incorrectly) that
> using tmpfs will make things faster.... if the buffer cache is doing
> it's job (ie you have enough memory) and you log the metadata
> transactions then tmpfs should not be any/much faster than real disk.

It depends on your access patterns. If you have a 30-50gb cache and
say, 1gb of RAM for caching, and 90% of your object requests come
from that 1gb of cached disk data, then you won't see the blocking
as being an issue.

But it doesn't work that way in reality. If it did, we wouldn't have
had to write aufs/diskd. :-)

Adrian
Received on Thu Aug 30 2001 - 18:47:06 MDT